2025 workshop program
A Diffusion-Based Machine Learning Model to Infer the Neutrino Effects from Large Scale Structure
The work
Abstract
Recent, and future, measurements of the large scale structure of the universe, such as DESI, Euclid, and the Nancy Grace Roman Space Telescope, have allowed us to measure the distribution of matter with increasingly higher precision. These measurements will allow us to map the distribution of matter within the universe at a higher resolution than ever before. This high resolution will allow us to investigate the neutrino mass, which suppresses structure formation on small scales. To extract the maximum amount of information from these measurements, we need computationally expensive numerical simulations. To speed up the process of running these simulations, we propose a diffusion-based generative machine learning model that will train on MG-PICOLA simulations. To validate our model, we use summary statistics such as the power spectrum, the bi-spectrum. This work will allow us to refine our theoretical models, especially at small, non-linear scales, where structure formation is driven by complex gravitational interactions.